Hong Kong’s AI twin stars are seeing renewed investor hype again: how far can the large-model commercialization stories of Zhipu and MiniMax go?

On July 15, 2026 (Beijing time), Hong Kong-listed AI stocks once again became the focus of market attention. The Hang Seng Index opened up 0.87%, and the Hang Seng Tech Index rose 0.78%. AI concept stocks saw a rebound across the board. Zhipu (02513.HK) opened up more than 6%, surged by over 5% at one point, but as of the time of publication its gain narrowed to 0.19%, trading at HK$1,603. MiniMax (00100.HK) rose 7.48%, to HK$247.2. The semiconductor sector strengthened in parallel: Semiconductor Manufacturing International Corporation (SMIC) rose 2.5% to HK$80.3, and Huali Semiconductor rose 2.6% to HK$175.9.

This move is not an isolated event. Since July 8, Zhipu and MiniMax have each gone through a series of key milestones—such as the release of restricted shares, large-scale refinancing, and an internal letter from their founders—resulting in significant divergence in their stock price performances. Zhipu jumped 13.85% against the trend on the lock-up release day, while MiniMax suffered a sharp single-day drop of nearly 18% after close to half of its shares were released. By the close on July 13, the gap in market capitalization between the two companies had already widened to more than tenfold.

The market is asking a deeper question: Are Hong Kong-listed AI large-model companies moving from a “valuation-driven hype” stage into a “commercial value validation” stage?

Global AI investment enters the second phase: shifting from compute infrastructure to models and applications

To understand the changes in how Hong Kong-listed AI stocks are priced, we first need to clarify the evolution of the global AI industry cycle.

Huatai Securities Overseas Technology Chief He Pianpian divided the AI industry into development stages using Huang Renzhuo’s “baseball nine-inning” theory: 2023 to 2025 is AI’s first inning, with the industry’s focus on training large models, where compute power centered on GPUs and HBM; 2026 officially ushers in the second inning, with the industrial main line switching to Agent intelligent agents and the rollout of low-latency inference scenarios. In a report released on July 14, Goldman Sachs further pointed out that the AI industry is moving toward multipolarization. In the second half, the core of competition is “effective output”—meaning the companies that generate the highest “useful output” per $1 of AI investment.

This shift in the industry cycle is directly reflected in the pricing logic of capital markets. Over the past three years, the core narrative of global AI investment has been a “compute arms race”—whoever has more GPUs and larger-parameter models gets a valuation premium. But entering 2026, this logic is being revised. The market is now asking: Can compute spending translate into sustainable revenue? Can model capability generate commercial value in specific scenarios?

In its mid-July outlook, UBS Securities explicitly listed “model capability, monetization, and Token ROI” as three major key themes for China’s AI model industry in the second half of 2026. This implies that the valuation anchor for Chinese AI companies is shifting from “technical narrative” to “commercial validation.”

As the only capital market globally that concentrates pure large-model listed companies, Hong Kong naturally becomes the most direct window to observe this valuation logic shift. Zhipu and MiniMax listed on the Hong Kong Stock Exchange in January 2026, and both were warmly welcomed by the market at launch. Zhipu’s issue price was HK$116.2, and MiniMax’s issue price was also met with substantial oversubscription. But half a year later, their stock price trajectories and market capitalization levels showed a fundamental split. This precisely reflects the market’s differentiated pricing for different paths to commercialization.

Zhipu and MiniMax: two distinctly different routes to AI commercialization

Although Zhipu and MiniMax are both dubbed Hong Kong’s “two champions” of large models, they differ fundamentally in technical route, market positioning, and commercialization strategy.

Zhipu (02513.HK): enterprise AI and China-native large-model ecosystem

Zhipu’s core narrative revolves around enterprise AI services and China-native large-model infrastructure. The company grew out of laboratories at Tsinghua University, and its GLM series models are seen as an important representative of domestic large models. As of July 2026, Zhipu has built a full-stack technology ecosystem spanning from chip adaptation to model deployment. It has open-sourced more than 60 cutting-edge models globally under MIT licensing; downloads in the international community have exceeded 100 million times; and it serves more than 5 million enterprise users and developers.

In terms of commercialization, Zhipu shows “pricing power” behavior. According to a research report from YuanDe Securities, Zhipu can double its API price while still sustaining continued business volume growth. China Yangtze Securities Group estimated that from 2026 to 2028, the company’s revenue would be RMB 2.5 billion, RMB 6.5 billion, and RMB 12.5 billion, respectively—up 244%, 162%, and 91% year over year. JPMorgan adjusted Zhipu’s target price up twice within a week, from HK$1,800 to HK$2,000 and then to HK$2,400, believing that its fund-raising will effectively ease the compute supply bottleneck.

On July 11, Zhipu founder Tang Jie published an internal letter announcing the launch of “Touch High.” It clarifies that in the next two years the company will strategically invest in AGI fundamental research rather than pursuing short-term application monetization. In the letter, he said: “Since the endpoint is AGI, then short-term benefits or industry hotspots are only the scenery along the way to the endgame.” This strategic choice was recognized by long-term capital. On the lock-up release day of July 8, Zhipu shares not only did not fall but rose 13.85%, delivering a rare “lock-up release-day celebration.”

MiniMax (00100.HK): multimodal AI and global C-end expansion

MiniMax took a different route. The company’s core strength lies in multimodal capabilities and a C-end product layout. Its flagship model, M3, is a native multimodal model that supports image and video input and desktop operations. In the application layer, MiniMax has built C-end products such as Talkie / Xingye (emotional interaction) and Hailuo AI (content creation), cumulatively serving over 200 million users globally, with daily average usage time for Talkie / Xingye users exceeding 70 minutes.

On its commercialization goals, MiniMax appears more aggressive. A Goldman Sachs research report on July 3 showed that management is confident about achieving an annual recurring revenue (ARR) target of $1 billion by the end of 2026.

However, the capital market’s valuation of MiniMax has been far less optimistic than that of Zhipu. On July 9, MiniMax saw approximately 150 million shares—representing 48.9% of total issued shares—released from lock-up. The float surged from under 3% to nearly 50%. On the day of release, the stock price crashed 17.98%, closing at HK$297.4. Since then, the price has continued to weaken. On July 14, it briefly fell to HK$209.2, a historical intraday low. By the close on July 13, MiniMax’s market capitalization had fallen from a peak of over HK$410 billion reached in March to under HK$70 billion.

On July 10, MiniMax founder Yan Junjie announced that effective immediately he would stop receiving any salary until the achievement of AGI, and he also put up shares equivalent to 4% of the company’s total issued share capital for team incentives. The same day, the company announced it would raise about HK$16.04 billion through a share placement and issuing convertible bonds. But these measures did not stop the decline. JPMorgan cut MiniMax’s target price twice within a week, from HK$300 to HK$240.

Two paths, two fates—reflecting a fundamental shift in how the capital market values China’s AI large-model companies.

Reconstructing the valuation logic: from user numbers and traffic to model capability and commercial revenue

The divergence in market capitalization between Zhipu and MiniMax fundamentally reflects the market’s shift in valuation anchors for AI companies.

Past valuation logic: user numbers + traffic

In the first phase of AI investment, the market focused more on user scale and traffic metrics such as DAU and call volume. MiniMax, with user scale and an overseas expansion story from C-end products such as Talkie / Xingye, once gained a very high valuation premium. On its first day of listing, shares surged 109%, and in March the share price climbed to HK$1,330, while market capitalization exceeded HK$410 billion. However, the traffic narrative failed to convert into sustained market confidence.

Future valuation logic: model capability + API revenue + Agent ecosystem + enterprise service revenue

Entering 2026, the market’s focus is undergoing a structural shift. According to an analysis by Securities Star (Zhengzhuo), the first batch of pure-play large-model listed companies in Hong Kong are seeing drastic valuation pullbacks. “As the market enthusiasm fades, capital is no longer paying a premium simply for ‘rare listed assets.’ The domestic large-model industry has officially bid farewell to the parameter narrative and the era of scarcity premiums. The valuation anchor is fully shifting to commercialization capability, compute cost control, and sustainable cash flows.”

More specifically, the new valuation framework at least includes four dimensions:

Model capability is the baseline threshold. Zhipu’s GLM-5 series led in the SWE-bench Pro coding evaluation, while GLM-5.2 topped the CodeArena evaluation to become the number-one globally available model. MiniMax’s M3 model also established a differentiated advantage in the multimodal domain. But model capability alone is no longer a sufficient condition—what the market cares about is whether these capabilities can translate into commercial revenue.

API revenue has become the most direct monetization validation metric. After doubling the API price, Zhipu still maintained business growth, demonstrating pricing power. MiniMax, meanwhile, expands its developer ecosystem through open-sourcing M3 model weights. The scale and growth rate of API revenue are becoming a key yardstick for measuring the commercial value of large-model companies.

Agent ecosystem is seen as the core form of next-generation AI applications. Zhipu lists “autonomous agent system” as one of the four technical engines in its “Touch High” plan; MiniMax’s Agent is positioned as a “general-purpose agent that can complete long-range complex tasks.” Huatai Securities pointed out that the industrial main line in 2026 has switched to deploying Agent intelligent agents. Whoever can build a sustainable Agent ecosystem first will seize the opportunity in the next stage.

Enterprise service revenue is becoming a key variable distinguishing valuation tiers. Zhipu’s enterprise-focused positioning aligns better with the current market style—an ecosystem serving more than 5 million enterprise users and developers, combined with the policy narrative of domestic substitution, makes it easier to obtain a valuation premium in today’s environment. By contrast, MiniMax’s C-end-dominant model faces higher pressures on user retention, paid conversion, and competition.

Tang Jingcao of Wumuxi Capital noted that the model layer is undergoing a major reshuffle: the current landscape of dozens of large-model companies coexisting in China is not sustainable. “As model capabilities converge, open-source ecosystems mature, and the big players’ free strategies continue, by 2027 the number of general-purpose large-model companies that can survive independently and develop smoothly is expected to be no more than 10.” Against this backdrop, capital markets will inevitably assess each company’s business model and sustainability using more stringent standards.

Value re-pricing or valuation normalization?

Returning to the question posed at the start of this article: are Hong Kong’s AI large-model companies entering a phase of value re-pricing?

The answer could be two-way. For Zhipu, the market is granting it a premium for “value re-pricing.” For MiniMax, the market is carrying out a harsh “valuation normalization” cycle.

Zhipu strengthened against the trend after the lock-up release, and institutions repeatedly raised their target prices, indicating that the market recognizes its “long-termism” strategy and its enterprise positioning. Data on July 14 showed that Northbound funds’ net purchases of Zhipu were HK$2.99B on a single day, and the continued inflow of long-term capital provides support for the share price. JPMorgan predicts that newly added inference compute resources for Zhipu could be converted into annual recurring revenue within the next 12 months.

MiniMax, however, faces a more complex challenge. The lock-up release radically changed the share supply structure—from under 3% float to nearly 50%—fundamentally altering its pricing logic. Combined with the pressure for financial investors to exit, the equity dilution effect brought by financing, and market doubts about the sustainability of its C-end business model, MiniMax’s valuation normalization may not be complete yet.

That said, it is worth noting that on July 14, CICC still maintained a “Buy” rating on MiniMax, believing that as a scarce resource in the AI large-model space, its revenue will continue growing rapidly and its infrastructure deployment pace is beyond expectations. This means the market’s pricing on MiniMax is not uniformly bearish; it is in the middle of a period of intense re-pricing.

From a more macro perspective, China’s AI large-model industry fundamentals have not deteriorated. According to OpenRouter data, weekly calls for China’s AI large models reached 23.45 trillion Tokens, up 15.01% month-over-month, and have remained at the top globally for ten consecutive weeks, ahead of the United States. The top six globally by call volume are China-based models. The industry’s high level of momentum contrasts sharply with the capital market’s violent fluctuations—this is a typical feature when the industry cycle switches from an “expectation-driven” mode to a “validation-driven” mode.

For investors, Hong Kong’s AI large-model sector is entering a stage that demands more selectiveness. The tailwinds from traffic stories and scarcity premiums are fading. In their place is more granular assessment of the efficiency of model capability-to-monetization, API revenue growth rates, the ability to build an Agent ecosystem moat, and the depth of enterprise services. The divergent stock performance between Zhipu and MiniMax may simply be the beginning of this value re-pricing process.

FAQ

Q1: What are the stock codes for Zhipu and MiniMax on Hong Kong’s market?

Zhipu’s Hong Kong stock code is 02513.HK. It was listed on January 8, 2026 at an issue price of HK$116.2. MiniMax’s Hong Kong stock code is 00100.HK. It was listed on January 9, 2026. Both are referred to by the market as Hong Kong’s “two large-model champions,” and are currently the only two pure-play large-model listed companies on the Hong Kong Stock Exchange.

Q2: Why did Zhipu’s share price rise after the lock-up release, while MiniMax crashed?

The core differences are in three areas: lock-up size—Zhipu released only 5.76% of its shares, while MiniMax released as much as 48.9%; shareholder structure—Zhipu’s released shares are mainly industrial capital and long-term institutions, whereas MiniMax includes many financial investors; and financing method—Zhipu’s placement at HK$1,588 won full subscription by institutions, while MiniMax’s placement combined with convertible bonds raised dilution concerns.

Q3: What is Zhipu’s “Touch High” plan?

On July 11, 2026, Zhipu founder Tang Jie published an internal letter announcing the launch of “Touch High.” Over the next two years, it will strategically invest in AGI fundamental research and not pursue short-term application monetization. The plan focuses on four technical directions: long-range tasks, autonomous agent systems, fully self-training, and extreme safety governance. The company also simultaneously completed a HK$31.4 billion share placement to reserve funds for the plan.

Q4: How does MiniMax’s business model differ from Zhipu’s?

MiniMax mainly focuses on multimodal AI and C-end applications, with products such as Talkie / Xingye and Hailuo AI. It has cumulatively served more than 200 million users globally. Zhipu focuses on enterprise AI services, serving more than 5 million enterprise users and developers, and building a China-native large-model ecosystem. Together, they represent the two paths to AI commercialization: C-end and B-end.

Q5: What is the core of the future valuation of Hong Kong-listed AI large-model companies?

The market valuation anchor is shifting from “user numbers + traffic” to “model capability + API revenue + Agent ecosystem + enterprise service revenue.” Whether the company can convert technical capabilities into sustainable commercial revenue, whether it can build an ecosystem moat in the Agent era, and whether compute investments can deliver positive returns will become the core variables determining valuation.